pre-text revisited: do we understand how writing prompts function?
DESCRIPTION
Pre-text revisited: Do we understand how writing prompts function?. Mark Chapman. Contents. The testing context Prompt effect literature Prompt categorization The dataset - essays and discourse variables Analyses of essays by prompt Significant differences in essays by prompt - PowerPoint PPT PresentationTRANSCRIPT
Pre-text revisited: Do we understand how writing prompts function?
Mark Chapman
1. The testing context2. Prompt effect literature3. Prompt categorization4. The dataset - essays and discourse variables5. Analyses of essays by prompt6. Significant differences in essays by prompt7. Implications for writing test design
Contents
• Writing test• Select 1 prompt from 2• 30 minutes to compose a response• Scored independently by 2 raters
• Effect of prompt wording on score awarded– Leu, Keech, Murphy & Kinzer, 1982– Brossell & Ash, 1984
• Effect of prompt wording on discourse variables– Greenberg, 1981 – Hirokawa & Swales, 1986
• Effect of different integrated prompts on discourse variables– Cumming et al., 2005 – O’Loughlin & Wigglesworth, 2007
Prompt effect in writing assessment
Problems with previous studies
• Used holistic score as measure of prompt effect
• Compared prompts that were very similar• Had small sample sizes• Analyzed stimuli differences not prompt • Little prompt categorization
1. What are the distinguishing characteristics of independent writing prompts?
2. How do these characteristics affect the test-takers’ final written products?
Research Questions
• Applied previous prompt taxonomies• Swales, 1982; Horowitz, 1991; Lim, 2009
Writing prompt characteristics
Prompt characteristic Examples
Domain Educational, occupational, personal, public
Response mode Argumentative or narrative
# of rhetorical cues 1 - 6
Open / focused Extent of contextualization or explicit guidance
The dataset
Prompts Responses per prompt
Responses by proficiency level
Discourse variables
6 60 20 15
Analyses
1. Factor structure of discourse variables (DVs)
2. Minimize number of MANOVA analyses to reduce risk of Type I error
3. MANOVA analyses of DV differences in responses elicited by each prompt
Factor Analysis
ComponentInitial Eigenvalues
Rotation Sums of Squared Loadingsa
Total % of Variance Cumulative % Total1 3.439 22.924 22.924 2.9122 2.027 13.510 36.435 2.3833 1.837 12.247 48.682 1.9644 1.486 9.907 58.588 2.0605 1.449 9.663 68.252 1.6856 1.070 7.134 75.386 1.2397 .841 5.606 80.992 1.3758 .744 4.962 85.953 9 .662 4.416 90.370 10 .491 3.276 93.645 11 .378 2.518 96.163 12 .302 2.011 98.175 13 .230 1.533 99.708 14 .037 .244 99.952 15 .007 .048 100.000
Extraction Method: Principal Component Analysis.
Factor structurePattern Matrixa
Component1 2 3 4 5 6 7
FREQ 1 .996 FREQ 3 -.805 FREQ 2 -.783 TTR -.358 CONCAUSi .847 CONLOGi .841 CONi .782 .414ERR/100 .978 TOTERR .932 AWL .919 FREQ AC .914 SYNLE .863 ASL .863 word # .991 CONTEMPi .978
MANOVA – lexical sophisticationMultivariate Testsa
Effect Value FHypothesis
df Error df Sig.Prompt Pillai's Trace .269 5.109 20.000 1416.00
0 .000
Wilks' Lambda .751 5.260 20.000 1165.085 .000
Hotelling's Trace .306 5.350 20.000 1398.000 .000
Roy's Largest Root .194 13.760c 5.000 354.000 .000
Tests of Between-Subjects Effects
SourceType III Sum of
Squares dfMean
Square F Sig.Prompt TTR .073 5 .015 3.262 .007
FREQ 1 1132.758 5 226.552 13.140 .000
FREQ 2 310.514 5 62.103 8.373 .000
FREQ 3 359.922 5 71.984 10.995 .000
MANOVA - cohesionMultivariate Testsa
Effect Value FHypothesis
df Error df Sig.
Prompt Pillai's Trace .232 5.946 15.000 1062.000 .000
Wilks' Lambda .776 6.222 15.000 972.118 .000
Hotelling's Trace .277 6.469 15.000 1052.000 .000
Roy's Largest Root .230 16.286c 5.000 354.000 .000
Tests of Between-Subjects Effects
SourceType III Sum of Squares df
Mean Square F Sig.
Prompt CONi 5956.622 5 1191.324 2.878 .015
CONCAUSi 3096.003 5 619.201 3.191 .008
CONLOGi17363.248 5 3472.650 12.865 .000
MANOVA - accuracyMultivariate Testsa
Effect Value FHypothesis
df Error df Sig.
Prompt Pillai's Trace .057 2.095 10.000 708.000 .023
Wilks' Lambda .943 2.103b 10.000 706.000 .022
Hotelling's Trace .060 2.111 10.000 704.000 .022
Roy's Largest Root .050 3.546c 5.000 354.000 .004
Tests of Between-Subjects Effects
SourceType III Sum of
Squares dfMean
Square F Sig.
Prompt TOTERR 1627.081 5 325.416 1.610 .157
ERR/100 67.973 5 13.595 .700 .624
MANOVA – academic vocabularyMultivariate Testsa
Effect Value FHypothesis
df Error df Sig.
Prompt Pillai's Trace .614 31.368 10.000 708.000 .000
Wilks' Lambda .472 32.111b 10.000 706.000 .000
Hotelling's Trace .933 32.856 10.000 704.000 .000
Roy's Largest Root .653 46.232c 5.000 354.000 .000
Tests of Between-Subjects Effects
SourceType III Sum of Squares df
Mean Square F Sig.
Prompt AWL 13.445 5 2.689 46.212 .000
FREQ AC 1129.167 5 225.833 31.165 .000
MANOVA – syntactic complexityMultivariate Testsa
Effect Value FHypothesis
df Error df Sig.
Prompt Pillai's Trace .194 7.611 10.000 708.000 .000
Wilks' Lambda .812 7.746b 10.000 706.000 .000
Hotelling's Trace .224 7.880 10.000 704.000 .000
Roy's Largest Root .182 12.896c 5.000 354.000 .000
Tests of Between-Subjects Effects
SourceType III Sum of Squares df
Mean Square F Sig.
Prompt SYNLE 115.864 5 23.173 10.222 .000
ASL 1066.851 5 213.370 10.200 .000
ANOVAs – fluency; writing score; temporal cohesive markers
ANOVA Sum of Squares df Mean Square F Sig.
word # Between Groups 125644.025 5 25128.805 3.399 .005
Within Groups 2616995.350 354 7392.642
Total 2742639.375 359
Writing score Between Groups 235.356 5 47.071 1.149 .334
Within Groups 14507.300 354 40.981
Total 14742.656 359
CONTEMPi Between Groups 4157.306 5 831.461 9.115 .000
Within Groups 32292.152 354 91.221
Total 36449.458 359
• Holistic score awarded
• Accuracy
Little evidence of significant differences in . . .
• Lexical sophistication• Cohesion• Academic language use• Syntactic complexity• Fluency
Significant differences in . . .
Prompts responsible for most differences
Prompt Domain Response mode Focus # of cues
4 Occupational Argumentative Focused 6
5 Personal Narrative Focused 6
6 Personal Narrative Open 2
• Score interpretation• Construct definition
• Consider which domains are appropriate• Consider permissible response modes• Consider number of rhetorical cues
Implications for writing assessment